Annonce

Les commentaires sont clos.

Internship subject : Deep Learning for super-resolution in remote-sensing hyperspectral imaging

21 December 2022


Catégorie : Stagiaire


I - Subject

Hyperspectral (HS) imaging [3] is now a well-established remote sensing modality for earth observation from satellites or planes. One of main characteristics of hyperspectral images is that they have several hundreds of spectral (colorimetric) bands – to be compared with 3 bands for classical RGB images. Therefore, for each pixel a full spectrum is acquired, enabling to extract some chemical properties of the materials present in the imaged scene. Nevertheless, such a precise spectral resolution is often obtained at the cost of a low spatial resolution, which is even more an issue when the HS sensor is far from the acquired scene, as it is the case in remote sensing.

We therefore propose in this internship to study the problem of HS super-resolution [2], which is a family of algorithms enabling to improve the (spatial) resolution of a given image [6, 5]. Beyond the classical iterative optimization-based methods, the aim of this internship will be to introduce new deep learning neural networks to tackle this problem. Specifically, several research paths are to be investigated :

  • Following our previous works, we propose to investigate the use of synthetic generative models to train HS super-resolution deep neural networks. Such a study is of particular interest, as quite few real HS data-sets are currently available. Specifically, we will first investigate the use deal leave models [1] ; the emphasis will be put on designing a precise model of the correlations exisiting between the HS spectral bands.
  • The design of interpretable neural networks for HS super-resolution. To do that, we propose to explore the applicability of Algorithm Unrolling (AU) [4] to the HS super-resolution problem. The main insight of AU is to design neural networks architectures mimicking the structure of conventional iterative algorithms : only a few parameters of such iterative algorithms are then learnt, requiring far less training samples than fully black-box neural networks. Compared to conventional iterative algorithms, AU-based neural networks are generally much faster at test time and they lead to better results, as they enable to have more data-driven approaches.

II - Candidate

The candidate must follow a Master 2 program (or equivalent) and have good knowledge of signal/image processing, as well as (deep) machine learning. Ideally, Python and its learning modules (Pytorch) should be known. In addition, knowledge about remote sensing is a plus, as well as in convex optimization.

During the internship, the candidate will acquire knowledge in image/signal processing, deep learning, convex optimization and inverse problems. The skills learnt can be useful in various domains : remote sensing, medical imaging, astrophysics...

 

III - Contact

The internship will be held in the IMAGES team (Télécom-Paris), under the supervision of Christophe Kervazo, Saïd Ladjal and Yann Gousseau.

Contact: christophe.kervazo@telecom-paris.fr;said.ladjal@telecom-paris.fr;yann.gousseau@telecom-paris.fr Possibility to continue as PhD : very likely.
More information on https://sites.google.com/view/christophekervazo/

 

IV - Références

[1] Raphaël Achddou, Yann Gousseau, and Sa ̈ıd Ladjal. Synthetic images as a regularity prior for image restoration neural networks. In International Conference on Scale Space and Variational Methods in Computer Vision, pages 333–345. Springer, 2021.

[2] Pierrick Chatillon, Yann Gousseau, and Sidonie Lefebvre. A statistically constrained internal method for single image super-resolution. In International Conference on Pattern Recognition, 2022.

[3] Nicolas Dobigeon, Jean-Yves Tourneret, Cédric Richard, José Carlos M Bermudez, Stephen McLaughlin, and Alfred O Hero. Nonlinear unmixing of hyperspectral images : Models and algorithms. IEEE Signal processing magazine, 31(1) :82–94, 2013.

[4] Karol Gregor and Yann LeCun. Learning fast approximations of sparse coding. In Proceedings of the 27th international conference on international conference on machine learning, pages 399–406, 2010.

[5] Wenming Yang, Xuechen Zhang, Yapeng Tian, Wei Wang, Jing-Hao Xue, and Qingmin Liao. Deep learning for single image super-resolution : A brief review. IEEE Transactions on Multimedia, 21(12) :3106–3121, 2019.

[6] Linwei Yue, Huanfeng Shen, Jie Li, Qiangqiang Yuan, Hongyan Zhang, and Liangpei Zhang. Image super- resolution : The techniques, applications, and future. Signal Processing, 128 :389–408, 2016.